D-separation

As explained in Causal AI

D-separation is a graphical criterion for determining whether conditioning on variables blocks all paths of statistical or causal association in a directed acyclic graph.

Suppose side-quest engagement is associated with in-game purchases. Does engagement cause purchases, or does guild membership influence both? A causal directed acyclic graph (DAG) represents these as different paths. D-separation checks which paths remain open after conditioning on selected variables, helping distinguish a causal route from a route through a common cause. section 7.2.4 section 4.1.2

Write U ⊥G V | Z to mean that node set Z d-separates U and V in graph G. Z is the conditioning set, also called the blockers. A path follows connected edges regardless of their arrow directions. U and V are d-separated only if every path between them is blocked; one open path makes them d-connected. The graphical claim is distinct from conditional independence in a probability distribution: it implies that independence if the DAG adequately represents the causal process. section 4.2.1 section 4.4.1

To check a path, inspect each middle node and the conditioning set:

• On a chain (i → m → j or i ← m ← j) or a common-cause fork (i ← m → j), conditioning on m blocks the path. section 4.2.1
• At a collider (i → m ← j), the path is blocked by default. Conditioning on m or one of its descendants can open it. Another node on that path may still block it. section 4.2.1 section 4.2.2

Figure 4.10 Colliders make d-connection tricky. Given a node m on a path, if m is not a collider, the path is d-connected by default and d-separated when you block on m. If m is a collider, the path is d-separated by default and d-connected when you block on m.

The diagram contrasts the default behavior of non-colliders and colliders when the middle node is conditioned on. section 4.2.1

A blocking set must cover every path

If U reaches V through U → I → V, U → J → V, and U → J → I → V, conditioning on {I, J} blocks all three. In a more complex graph, blocking one path can open another when the chosen blocker is a collider there. Recheck every path after adding a blocker; working from shorter paths toward longer unblocked ones can simplify the search. section 4.2.2

Listing 4.1 D-separation analysis of the DAG in figure 4.15

from networkx import is_d_separator      #1
from pgmpy.base import DAG     #2
dag = DAG([     #2
    ('I', 'U'),     #2
    ('I', 'M'),     #2
    ('M', 'U'),  #2
    ('J', 'V'),    #2
    ('J', 'M'),     #2
    ('M', 'V')     #2
])     #2
print(is_d_separator(dag, {"U"}, {"V"}, {"M"}))     #3
print(is_d_separator(dag, {"U"}, {"V"}, {"M", "I", "J"}))    #4
print(is_d_separator(dag, {"U"}, {"V"}, {"M", "I"}))     #5
print(is_d_separator(dag, {"U"}, {"V"}, {"M", "J"}))    #5

The Python listing constructs a DAG with pgmpy and uses NetworkX is_d_separator to compare candidate conditioning sets. section 4.2.3

D-separation is useful for both identification and model criticism. An adjustment set should block every backdoor path between treatment and outcome; an intervention represented by graph surgery instead removes incoming edges to its target. Conversely, if a proposed DAG d-separates variables that appear dependent in data, that evidence can refute the DAG. A statistical independence test can also err, so a failed or passed test is not an assumption-free verdict. section 11.3.1 section 4.3 section 4.4.1

How Causal AI covers it

How deep it goes

Defines path-level blocking rules, works through multiple paths and collider side effects, and checks candidate sets in Python.

Examples

Uses chain and collider diagrams, multi-path blocking sets, a gaming intervention, and a NetworkX check of a pgmpy DAG.

What it assumes

Assumes a causal DAG and conditioning variables; defines the path and collider rules needed to assess graphical separation.

What you can do after

Apply it via NetworkX and pgmpy to check conditioning sets and challenge DAG assumptions before causal estimation.

How it connects

Links causal DAGs and conditional independence to confounding, backdoor adjustment, graph surgery, and causal discovery.

liveBookmanning.com

Read Causal AI on liveBook Get the book at manning.com

Also explained in

• 

  liveBookmanning.com
  Causal Inference for Data ScienceD-separation is a graphical criterion stating that a conditioning set blocks the relevant paths between variables in a graph, implying their conditional independence under the graph's assumptions.

Compare how 2 books explain d-separation →

Related concepts

• conditional independencecontrasts with · Two variables are conditionally independent given specified information when their conditional joint distribution factors into the product of their conditional distributions.
• directed acyclic graphuses · A directed acyclic graph is a graph with directed edges and no directed cycles.
• graph surgeryused by · Graph surgery is the modification of a causal graph to represent an intervention by changing or removing the intervened variable's incoming causal edges.
• causal graphical modelproduced by · A causal graphical model is a probabilistic or structural model whose graph represents causal relationships and whose mechanisms support reasoning about interventions.
• backdoor criterionpart of · The backdoor criterion is a graphical condition for identifying a causal effect by requiring a variable set to block every backdoor path while excluding treatment descendants.
• statistical independence testuses · A statistical independence test is a hypothesis-testing procedure that evaluates whether variables are statistically independent or conditionally independent based on observed data.

This explainer was generated by Manning's knowledge graph from the text of Causal AI. The section links point to the parts of the book it draws on.